mmap or something else???....

classic Classic list List threaded Threaded
5 messages Options
Reply | Threaded
Open this post in threaded view
|

mmap or something else???....

Leandro Franco
Hi guys...


I'm having an error in a FUSE filesystem when I try to compile gcc...

The funny thing is that the error happens when I call the file system
demon from a shell but when I start it from anjuta (as far as i know
it's just a front end to gdb) it doesnt kill the file system, it just
says that mmap is not there.....


I dont really know if it's mmap, gcc or anything else in my code but
I'm kind of clueless and the error is quite annoying so any ideas are
really welcome....

i'm attaching a file with the make output of gcc for both cases.....


Thank you all for your time...


Leandro Franco

error-mmap (6K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: mmap or something else???....

Miklos Szeredi
> I'm having an error in a FUSE filesystem when I try to compile gcc...
>
> The funny thing is that the error happens when I call the file system
> demon from a shell but when I start it from anjuta (as far as i know
> it's just a front end to gdb) it doesnt kill the file system, it just
> says that mmap is not there.....

What's the FUSE version?  Is there any Oops in dmesg output?

Try setting the core-dump limit to infinity:

   ulimit -c unlimited

Then run the filesystem.  And after it crashes, look at the corefile
with gdb (or anjuta).

> I dont really know if it's mmap, gcc or anything else in my code but
> I'm kind of clueless and the error is quite annoying so any ideas are
> really welcome....

Well, mmap is not fully supported in FUSE.  Only read-only or private
mappings work.  The configure script is probably checking for
writeable mmap, and it's finding that it doesn't work.

It's actually a bit stupid for gcc to test this at compile time, since
at runtime it could be running on a different filesystem which _does_
fully support mmap.  You could take up the issue with the gcc people,
but it doesn't sound to be a serious problem.

> i'm attaching a file with the make output of gcc for both cases.....

A bit more of the debug trace would have been useful.

At the end of the trace there's an UNLINK and then a RELEASE for
probably the same file.  This could be the cause of the problem.

What mount options do you give to FUSE?  Is 'hard_remove' one of them?

Thanks,
Miklos


-------------------------------------------------------
This SF.Net email is sponsored by Yahoo.
Introducing Yahoo! Search Developer Network - Create apps using Yahoo!
Search APIs Find out how you can build Yahoo! directly into your own
Applications - visit http://developer.yahoo.net/?fr=offad-ysdn-ostg-q22005
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel
Reply | Threaded
Open this post in threaded view
|

Re: mmap or something else???....

Leandro Franco
Hi again...

after the last email I tried the 'hard_remove' option and the problem
went away....
the gcc test even says that mmap works.....

I dont know the consecuences but for the moment i dont need to delete
open files and such things in the file system so i will reluctantly
move on.....

but if you have any ideas.... they are always welcome....


Thank you very much for your effort

Leandro Franco


> thanks for the help and the quick reply ....
>
> > What's the FUSE version?  Is there any Oops in dmesg output?
>
> 2.2 .... and I havent seen any...
>
> > Try setting the core-dump limit to infinity:
> >
> >    ulimit -c unlimited
> >
> > Then run the filesystem.  And after it crashes, look at the corefile
> > with gdb (or anjuta).
>
> ok, I will try... for the moment i would like you to take a look of
> debug output....
>
> now i'm really lost because it mentions the fuse_hidden files.... (the
> possibility of my doing something silly is significant... it could be
> a really trivial error but for the moment i dont see it)...
>
> and the last call to the hidden file is the one after crashing the file system
>
> unique: 1325, opcode: LOOKUP (1), nodeid: 10, insize: 49
> LOOKUP /src/gcc-3.4.3/intl/conftestmmap
>    unique: 1325, error: -2 (No such file or directory), outsize: 16
> unique: 1326, opcode: MKNOD (8), nodeid: 10, insize: 57
> MKNOD /src/gcc-3.4.3/intl/conftestmmap
> mknod   : /src/gcc-3.4.3/intl/conftestmmap
>    NODEID: 61
>    unique: 1326, error: 0 (Success), outsize: 136
> unique: 1327, opcode: OPEN (14), nodeid: 61, insize: 40
> open : /src/gcc-3.4.3/intl/conftestmmap
> OPEN[154812024] flags: 0x1
>    unique: 1327, error: 0 (Success), outsize: 28
> unique: 1328, opcode: WRITE (16), nodeid: 61, insize: 4156
> WRITE[154812024] 4096 bytes to 0
> write : /src/gcc-3.4.3/intl/conftestmmap
>    unique: 1328, error: 0 (Success), outsize: 20
> unique: 1329, opcode: RELEASE (18), nodeid: 61, insize: 48
> RELEASE[154812024] flags: 0x1
> release : /src/gcc-3.4.3/intl/conftestmmap
>    unique: 1329, error: 0 (Success), outsize: 16
> unique: 1330, opcode: OPEN (14), nodeid: 61, insize: 40
> open : /src/gcc-3.4.3/intl/conftestmmap
> OPEN[154812024] flags: 0x2
>    unique: 1330, error: 0 (Success), outsize: 28
> unique: 1331, opcode: READ (15), nodeid: 61, insize: 56
> READ[154812024] 4096 bytes from 0
> read : /src/gcc-3.4.3/intl/conftestmmap
>    READ[154812024] 4096 bytes
>    unique: 1331, error: 0 (Success), outsize: 4112
> unique: 1332, opcode: UNLINK (10), nodeid: 10, insize: 49
> UNLINK /src/gcc-3.4.3/intl/conftestmmap
>    unique: 1332, error: 0 (Success), outsize: 16
> unique: 1333, opcode: RELEASE (18), nodeid: 60, insize: 48
> RELEASE[0] flags: 0x0
> release : /src/gcc-3.4.3/intl/conftest
>    unique: 1333, error: 0 (Success), outsize: 16
> unique: 1334, opcode: RELEASE (18), nodeid: 61, insize: 48
> RELEASE[154812024] flags: 0x2
> release : /src/gcc-3.4.3/intl/.fuse_hidden0000003d00000001
> unlink  : /src/gcc-3.4.3/intl/.fuse_hidden0000003d00000001
>
> and from gcc:
> checking for working mmap... rm: cannot remove `conftest*': Transport
> endpoint is not connected
>
>
> I suppose it deletes the file while it's still open so FUSE creates
> the hidden file... but somewhere between closing it and deleting it my
> system crashes.....
>
>
> > A bit more of the debug trace would have been useful.
> >
> > At the end of the trace there's an UNLINK and then a RELEASE for
> > probably the same file.  This could be the cause of the problem.
> >
> yeap.... i think it's the same file.... but it has been renamed.... i
> dont know the details of FUSE so I'm not sure if I'm doing something I
> shouldnt
>
> > What mount options do you give to FUSE?  Is 'hard_remove' one of them?
> >
> just like that:
> -d -o allow_other,large_read
>
> and....
> Linux 2.4.21-27.0.2.EL
>
> > Thanks,
> > Miklos
> >
>
> Thank you....
>
> Leandro Franco
>


-------------------------------------------------------
This SF.Net email is sponsored by Yahoo.
Introducing Yahoo! Search Developer Network - Create apps using Yahoo!
Search APIs Find out how you can build Yahoo! directly into your own
Applications - visit http://developer.yahoo.net/?fr=offad-ysdn-ostg-q22005
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel
Reply | Threaded
Open this post in threaded view
|

Re: mmap or something else???....

Miklos Szeredi
> after the last email I tried the 'hard_remove' option and the problem
> went away....
> the gcc test even says that mmap works.....

Hmm. That is strange indeed.

Can you try 2.3-rc1 without the 'hard_remove' option?  Does it happen
too?

Thanks,
Miklos


-------------------------------------------------------
This SF.Net email is sponsored by Yahoo.
Introducing Yahoo! Search Developer Network - Create apps using Yahoo!
Search APIs Find out how you can build Yahoo! directly into your own
Applications - visit http://developer.yahoo.net/?fr=offad-ysdn-ostg-q22005
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel
Reply | Threaded
Open this post in threaded view
|

Re: mmap or something else???....

Leandro Franco
mmm...

the same behavior :/


Thank You,

On 6/2/05, Miklos Szeredi <[hidden email]> wrote:

> > after the last email I tried the 'hard_remove' option and the problem
> > went away....
> > the gcc test even says that mmap works.....
>
> Hmm. That is strange indeed.
>
> Can you try 2.3-rc1 without the 'hard_remove' option?  Does it happen
> too?
>
> Thanks,
> Miklos
>


-------------------------------------------------------
This SF.Net email is sponsored by Yahoo.
Introducing Yahoo! Search Developer Network - Create apps using Yahoo!
Search APIs Find out how you can build Yahoo! directly into your own
Applications - visit http://developer.yahoo.net/?fr=offad-ysdn-ostg-q22005
_______________________________________________
fuse-devel mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/fuse-devel